Prevalence of BRCA mutations among hereditary breast and/or ovarian cancer patients in Arab countries: systematic review and meta-analysis

This present study was conducted following the recommendations provided by the PRISMA [9] and MOOSE guidelines [10]. This study was registered with the International Prospective Register of Systematic Reviews (PROSPERO) and the registration number is: CRD42018095905.

The following inclusion criteria were used to identify eligible studies: peer reviewed studies published in English or Arabic; cohort studies, cross-sectional, case-series or registry based studies, providing sufficient data to compute the prevalence of either BRCA1 or BRCA2 gene mutations in the breast and / or ovarian cancer patients in the Arab world.

Studies were excluded if they were confined mainly to the carriers of BRCA mutations without including other cancer patients. Those studies that has no information on family history of disease were also excluded as there is no way of quantifying HBOC. Likewise, experimental studies that were evaluating the effect of an intervention on HBOC patients were also excluded as computing prevalence in these studies were not possible. Additionally, studies reporting on BRCA1/2 genes mutations in cancers other than breast or ovarian cancers were also excluded.

The electronic databases Web of Science, PubMed/Medline and EMBASE were searched to identify the primary studies up to April 11, 2018. Further search included Google Scholar search to identify any relevant studies as well hand search of the reference lists of the relevant articles. Although, there were no systematic reviews or meta-analyses from this region, we searched for all review articles reporting on the BRCA mutations form other parts of the world to identify any relevant studies from Arab countries. Only those published in English and Arabic were included, and the duplicates studies were removed using EndNote software.

Studies from all 22 Arab member states of the Arab league [11] were included. We used the following terms to be in the “Titles” or “Abstracts: [BRCA],” and [“Arabs,” or “Qatar,” or “Kuwait,” “Oman,” or “Iraq,” or “Saudi Arabia,” or “Bahrain,” or “United Arab Emirates,” or “Yamen,” “Syria,” or “Lebanon,” or “Jordan,” or “Palestine,” or “Jerusalem,” or “Gaza,” or “Egypt,” or “Libya,” or “Sudan,” or “Tunisia,” or “Morocco,” or “Algeria,” or “Somalia,” or “Comoros,” or “Djibouti,” or “Mauritania.”] and any other variant names for any of these countries.

Two review authors independently assessed titles and abstracts of all citations retrieved by the search for relevance against the inclusion criteria. Then the full-text versions of studies considered potentially eligible were retrieved. The same two authors independently assessed the full papers for eligibility, with disagreements resolved through input by a third author [12]. When the eligibility of a study was unclear, we attempted to contact study authors.

Two authors independently extracted data from eligible studies and cross-checked for accuracy and agreement by the third author. Data extraction using a standardized form included variables such as author, year of publication, country, city, study design, type of disease (BC/OC), age at diagnosis, mutation type, mutation test used, overall sample size and sample size of HBOC subset patients (Table 1).

We used the risk of bias assessment tool developed by Hoy et al. [13] to evaluate the quality of individual studies. The tool has 10 items: four items were related to external validity in terms of how well study sample represent the study population and remaining six items were related to internal validity of the studies. Total quality scores for each study was obtained by summing the binary response scores (Yes = 1 or No =0) that ranged from 0 to 10. As per Hoy et al. [13], scores above 7 indicated low risk of bias while a score between 6 to 7 indicated moderate risk of bias. Those studies scoring less than 6 were considered to have high risk of bias.

We pooled the quantitative data using Meta-XL version 5.3 [14]. We reported the pooled prevalence and 95% confidence intervals (CI) and explored the robustness of meta-analyses using appropriate meta-analytical models based on the level of heterogeneity. Statistical heterogeneity were assessed using Cochrane Q statistic and I-squared (I²) statistics [15].

In the absence of heterogeneity, the Fixed Effect Models (FEM) were used. When substantial heterogeneity is encountered, the Quality Effects Models (QEM) were used to pool the prevalence [16]. Significant heterogeneity is indicated by a significant P-value of the Cochrane-Q test or an I² statistic value that is higher 50%. We used the quality scores computed by ROB assessment tool to fit the bias adjusted QEM. Additionally, Random Effect Models (REM) were used to obtain sensitivity estimates when substantial heterogeneity is encountered. We further investigated the sources of heterogeneity using subgroup analysis. We used a-priori defined test to compare the pooled prevalence in three different regions of the Arab world: Levant, North Africa and Arabian Gulf Co-operation Countries (GCC) using Z- tests for proportions.

The database search generated a total of 6315 records (Fig. 1). We excluded 6249 studies and reasons for exclusions are listed in the Fig. 1. We assessed the full text of 66 studies for their eligibility. Three authors independently read through the full texts and identified those eligible for the review based on a-priori inclusion criteria. Only 14 publications were found to be eligible and included in this study.

The 14 studies selected were all peer reviewed full length publications form nine different Arab countries. All these 14 articles were published in English and published during the period of 2007 and 2018. They reported on the prevalence of BRCA1 and / or BRCA2 mutations among breast or/ and ovarian cancer patients. Of these, eight were conducted in North African countries, four from Levant, and two from GCC. There were three studies from Morocco: one each from North-east Morocco [27], Casablanca [22] and Rabat [4]. Two studies each from the following countries. Algeria [23, 25], Lebanon [7, 26], and Tunisia [20, 21]. Single studies from each of the following countries: Qatar [17], Saudi Arabia [5], Jordan [24], Palestine [19], and Egypt [18].

Sample sizes of the included studies ranged between10 to 192, and total number of patients pooled was 917. Of these, 11 studies included Hereditary Breast Cancer (HBC) patients, three had Hereditary Breast and/or Ovarian Cancer (HBOC) patients, and a single study included only Heredity Ovarian Cancer (HOC) patients. Seven studies investigated both BRCA1 and BRCA2 gene mutations, six studies had data on only BRCA1 gene mutation and one study was exclusively on BRCA2 mutations. There were variation in the methods used to detect BRCA mutations. Three studies used direct sequencing, while others used varying types of methods.

Of the 14 articles included in our study, eight studies were found to have moderate risk of bias [4, 17‐23], and six had lower risk of bias [5, 7, 24‐27]. In term of the external validity components, only three articles were considered to have a good external validity [5, 25, 27] (See Fig. 2).

From studies reported total BRCA mutation cases, the overall pooled prevalence of BRCA among breast and/or ovarian cancer based on QEM was 20% (95% CI: 7–36%) [4, 5, 7, 17‐27] (Fig. 3a). The pooled prevalence of BRCA1 mutation among hereditary breast and/ or ovarian cancer was 12% (95% CI: 4–21%) (Fig. 3b) [4, 5, 7, 17, 19‐25]. The estimated pooled prevalence of BRCA2 was 12% (95% CI: 4–25%) based on five studies reported such mutation in the HBOC population [7, 17, 21, 22, 27] (Fig. 3c). Further, six studies reported prevalence of BRCA mutation among non- hereditary sample [4, 5, 7, 18, 22, 23]. The pooled prevalence in this population was estimated to be 10% (95% CI: 0–28%).

Of all 14 studies included, 13 reported prevalence of BRCA among breast cancer patients and the pooled prevalence was 21% (95% CI: 6–40%) [4, 5, 7, 17, 19‐26]. Of this, the prevalence of BRCA1 mutation alone among breast cancer patients was 11% (95% CI: 4–21%) [4, 5, 7, 17, 19‐25]. Likewise, the prevalence of BRCA2 mutations among breast cancer patients was 17% (95% CI: 9–27%) [7, 17, 21, 22, 24]. Only one study reported the prevalence of BRCA1 mutations among ovarian cancer patients which was in Algeria [25] with a prevalence of 1% (95% CI: 0–2%).Three studies [17, 22, 27] out of 14 studies report a prevalence of BRCA1 or BRCA2 among both Breast and Ovarian Cancer patients, the pooled prevalence was 9% (95% CI: 0–25%).

All the estimates provided above are QEM based to adjust for bias in individual studies. However, the Random Effect Model (REM) is a commonly used and is recommended by Cochrane group as the choice of model when there is substantial heterogeneity between studies included in a meta-analysis [12]. We therefore used REM as a sensitivity method to compare the QEM estimates. Table 2 shows a comparison between QEM and REM estimates, which were not much different.

We repeated the primary analysis of overall pooled prevalence of BRCA among hereditary breast or/and ovarian cancer by only including the low risk of bias studies to estimate the prevalence of high quality studies [5, 7, 24‐27]. The pooled prevalence of BRCA mutations among the high quality studies were estimated to be 11% (95CI: 1–27%). This compared to the overall pooled prevalence including all studies was lower and much more comparable with the global estimates.

Quality effect model (QEM) used for subgroup analysis from different geographical locations in GCC, Levant and North Africa to understand any geographic variations in the prevalence of BRCA mutations among hereditary breast or/and ovarian cancer patients. The highest pooled prevalence of BRCA mutations was estimated in Levant region the pooled prevalence was 28% (95% CI: 11–49%) followed by the Gulf region where the pooled prevalence was 22% (95% CI: 0–58%). Comparable prevalence in the North African region was lower than other regions, as it was estimated to be 16% (95%CI: 0–44%). These estimates did not differ statistically based on Z- test for proportions.

Significant heterogeneity was encountered throughout the analysis as indicated by the p values of the Cochrane Q statistics and I² statistics values. For instance, the studies [4, 5, 7, 17‐27] included to pool the prevalence for BRCA mutations had a Cochran Chi-square p value of < 0.001 and I² of 95%. Likewise, there were significant heterogeneity across the studies used to obtain the pooled prevalence for BRCA1 among total HBOC (Chi-square = 69.2, p-value< 0.001, I² = 86%) [4, 5, 7, 17, 19‐25]. For BRCA2 heterogeneity between the studies were also substantial (Chi-square = 23.5, p-value < 0.001, I² = 79.7%) [7, 17, 21, 22, 27].

For total BRCA mutations among general population, there was a considerable heterogeneity across studies (Chi-square = 70, P value = 0.001, I- square = 94.3%) and the pooled prevalence was 35% (95% %CI: 16–55%). The pooled prevalence of BRCA1 mutation among Breast Cancer Patients was 11% (95% CI: 4–21%) [4, 5, 7, 17, 19‐25], and there was a substantial heterogeneity across the studies (Chi-square = 79.3, P value < 0.001, I-square = 87.38%). As for the prevalence of BRCA 2 mutation among Breast Cancer patients, data was reported from 5 studies [7, 17, 21, 22, 24]. Where the pooled prevalence was 17% (95% CI: 9–27%), and significant heterogeneity was found (Chi-square = 18.5, P value = 0.001, I square = 78.39%). Only one study [25] reported BRCA1 among ovarian cancer and no studies reported BRCA2 among ovarian cancer patients, so heterogeneity was not assessed. For BRCA1 or/and BRCA2 among patients with both Breast and Ovarian cancer, the pooled prevalence was 9% (95% CI: 0–25%) [17, 22, 27]. For heterogeneity, there was large heterogeneity detected (Chi-square = 19, P value < 0.001, I-square = 89.4%).

Publication bias was investigated using classical funnel plot, where an asymmetry indicated potential publication bias (Fig. 4a, b and c). As for BRCA1 and BRCA2 mutations estimates in hereditary patients, funnel plots showed a potential publication bias. However, Hunter et al. [28] suggested that conventional funnel plots, that assess publication bias, are inaccurate for prevalence meta-analysis. We are unsure about the asymmetry observed in our study if it is true reflection of publication bias or simply an over-estimate as argued by Hunter et al. [28].